Deep Gallery: A Convolutional Neural Network Algorithm of Artistic Style Transfer

This project requires:

• numpy, scipy
• tensorflow
• cv2
• python2.7

To create an Deep-Gallery conda environment:

conda env create -f conda_env.yml

1. Download pre-trained vgg19 and coco train2014 dataset:

wget http://www.vlfeat.org/matconvnet/models/beta16/imagenet-vgg-verydeep-19.mat
wget http://msvocds.blob.core.windows.net/coco2014/train2014.zip
unzip train2014.zip

Make sure imagenet-vgg-verydeep-19.mat and train2014 are under in Deep-Gallery/.

2. Set parameters in main.py:

STYLE_PATH = './wave.jpg'
TEST_PATH = './artist.jpg'

3. Start to build a model:

python main.py --function train

1. Set parameters in main.py:

CONTENT_PATH = './input/artist.jpg'
MODEL_PATH = './8ca14295/ck_dir/model_2500.ckpt'
GENRD_PATH = './output/artist.jpg'

2. Start to transfer a picture:

python main.py --function transfer

python main.py --function transfer --reserve

This the last project assignment in 10701. As the saying goes, Keep calm and trust the process. All assigments in this class are very struggle, but after mindful thinking and continuous trials, we have learnt a lot. So thank all of the faculty members, this is the best Machine Learning courses we have token in CMU!